Multi-domain transfer component analysis for domain generalization

T. Grubinger, A. Birlutiu, H. Schöner, T. Natschläger, T. Heskes. Multi-domain transfer component analysis for domain generalization. Neural Processing Letters, pages online first, DOI 10.1007/s11063-017-9612-8, 4, 2017.

  • Thomas Grubinger
  • Adriana Birlutiu
  • Holger Schöner
  • Thomas Natschläger
  • Tom Heskes
JournalNeural Processing Letters
Seitenonline first

This paper introduces a framework for domain generalization, called Multi-TCA (Multi-Domain Transfer Component Analysis). Domain generalization addresses the question of how to use the knowledge acquired from related domains on new domains. Multi-TCA is based on Transfer Component Analysis (TCA) which is a popular transfer learning technique. TCA learns a shared subspace by minimizing the dissimilarities across domains, while maximally preserving the data variance. We introduce Multi-TCA which is an extension of TCA to multiple domains as well as Multi-SSTCA, which is an extension of TCA for semi-supervised learning. Multi-TCA and Multi-SSTCA are evaluated and compared with other state-of-the-art methods on real and simulated data. Experimental results demonstrate that Multi--TCA can improve predictive performance on previously unseen domains.